259. VARIMA
Vector ARIMA
Differencing applied component-wise to each of the series.
In practice, cointegrated systems use VECM (Vector Error Correction) rather than VARIMA — VECM models the long-run equilibrium directly.
Parameters: , , ,
Orders: , , ,
Example:
Given
- Orders: , , ,
-
Coefficient matrices:
- Intercept:
- Initial conditions: ,
- Two series stacked into :
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | |
| 12 | 10 | 8 | 11 | 14 | 12 | 9 | 13 | 16 | 14 | 11 | 15 | 18 | 16 | 13 | 17 |
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | |
| 8 | 9 | 7 | 6 | 10 | 11 | 9 | 8 | 12 | 13 | 11 | 10 | 14 | 15 | 13 | 12 |
Step 1 — formula
Substitute orders into the VARIMA recursion:
Difference — apply component-wise:
In differenced space, the model is VARMA:
Forecast the difference (set ):
Undifference — convert back to a forecast for :
Innovation:
Step 2 — apply at
First difference: .
Plug in , , , :
Step 3 — iterate
Pipeline at each : difference VARMA-forecast undifference. Values rounded to 4 decimal places.
| 2 | ||||
| 3 | ||||
| 4 | ||||
| 5 | ||||
| 6 | ||||
| 7 | ||||
| 8 | ||||
| 9 | ||||
| 10 | ||||
| 11 | ||||
| 12 | ||||
| 13 | ||||
| 14 | ||||
| 15 | ||||
| 16 |